Cardiovascular Health and Disease Across the Life Course

Many Roads to Rome

Sarah Urbut, MD PhD

Massachusetts General Hospital

2025-10-17

The Life Course Paradigm Shift

Current Risk Models: The 10-Year Problem

Traditional Approach:

  • Pooled Cohort Equations (PCE)
  • 10-year ASCVD risk
  • Treatment at thresholds (≥7.5%)

The Question: > “Who will have an event in the next 10 years?”

What’s Missing:

  • Young patients with high lifetime risk
  • Low short-term but high long-term risk
  • Different disease trajectories
  • Optimal intervention timing

Better Question: > “What is their lifetime burden, and when should we intervene?”

Many Roads to Rome

Key Insight

Traditional risk models assume one pathway. Reality is far more complex—different trajectories require different intervention strategies.

CV Health Personas

Persona 1: High Genetic Risk

Clinical Profile:

  • High polygenic risk score (PRS)
  • Or familial hypercholesterolemia (FH)
  • LDL 180 mg/dL at age 35
  • CAD onset by age 50

Research Evidence:

  • Ference et al., NEJM 2012: Genetic variants that lower cholesterol even moderately over lifetime dramatically reduce CAD
  • Khera et al., Nat Genet 2018: PRS identifies high-risk individuals missed by traditional screening

Your Model Findings:

Signatures change 5-10 years before clinical diagnosis

Persona 2: Metabolic Trajectory

Progression Path:

  1. Prediabetes (age 40-45)
  2. Type 2 diabetes (age 45-50)
  3. Metabolic syndrome (age 50-55)
  4. CAD (age 55-60)

Key Research Finding:

Urbut et al., medRxiv 2024: > “Metabolic signature (Sig 12) rises first, cardiovascular signature (Sig 5) follows 2-3 years later”

The Intervention Window: - When metabolic signature plateaus - But CVD signature accelerates - 2-3 years before clinical CAD

Signature Transition:

Persona 3: Inflammatory Pathway

Clinical Picture:

  • Rheumatoid arthritis, psoriasis, or lupus
  • Often normal lipids
  • “Low” 10-year PCE risk
  • But ongoing systemic inflammation

Your Research:

  • 52 patients: rheumatologic → CV transitions
  • Signature deviations 5 years pre-CAD
  • Multi-signature interactions:
    • Metabolic + Inflammatory together
    • Synergistic effect (9.1% > 4.3% + 2.8%)

Evidence Base:

  • ACC Statement on Inflammation: Inflammation is therapeutic target in CVD
  • CANTOS trial: IL-1β inhibition reduces CV events

Signature Behavior:

Often missed: Normal lipids but high inflammatory burden

Persona 4: Healthy Ager

Characteristics:

  • Low signature loadings across decades
  • Maintained cardiovascular health
  • Protective factors?
    • Genetics
    • Lifestyle
    • Both?

Model Performance:

  • Baseline AUC: 0.898 (ASCVD)
  • 1-year washout: 0.701
  • 2-year washout: 0.680

Key Question: What distinguishes this group? Can we replicate their trajectory in others?

Clinical Impact

Same 10-year risk ≠ Same intervention. Trajectory determines timing and intensity of prevention.

Dynamic Risk: Life Course Models

The Aladynoulli Model

What It Does:

  • Longitudinal disease signatures over 30+ years
  • Bayesian survival framework
  • Dynamic risk updates

What It Incorporates:

  1. Genetics (fixed): Polygenic risk scores
  2. Biomarkers (dynamic): Serial lipids, BP, glucose
  3. Imaging (when available): CAC, plaque burden
  4. Events (transitions): New diagnoses as signals

Scale:

  • 348 diseases
  • 400,000 individuals (UK Biobank)
  • 21 disease signatures
  • 50+ time points (ages 30-80)

Mathematical Framework:

\[\pi_{i,d,t} = \kappa \times \sum_k \theta_{i,k,t} \times \phi_{k,d,t}\]

Where:

  • \(\theta_{i,k,t}\) = signature proportions (from Gaussian process)
  • \(\phi_{k,d,t}\) = disease probabilities
  • \(\lambda_{i,k,t} \sim GP(\mu_k + G_i\gamma_k, K_\lambda)\) (temporal dynamics)

Key Innovation:

Signature velocity (rate of change) predicts risk:

Important

“Top quartile Sig 5 velocity has HR=2.8 for CAD”

This translates to 18 months earlier CAD onset

Novel Biomarker: Signature Velocity

Traditional Approach:

  • Single time point measurement
  • Example: LDL = 150 mg/dL

Life Course Approach:

  • Trajectory and velocity
  • Example: LDL 120→180 over 10 years
  • Rate of change = 6 mg/dL/year

Your Discovery:

“Fast signature progression → higher disease risk independent of absolute level”

Clinical Translation:

  • Top quartile velocity: HR = 2.8 for CAD
  • 18 months earlier onset
  • New therapeutic target: Slow progression

Example: Same Risk, Different Trajectories

Two 55-year-old women, both with 10-year ASCVD risk of 7.5%:

Feature Woman A Woman B
LDL trajectory Stable 130 mg/dL × 20 years 120→180 mg/dL over 15 years
Inflammation Recent hsCRP uptick (1→4 mg/L) Stable CRP < 2 mg/L
Signature pattern Inflammatory sig rising Metabolic sig dominant
PCE risk 7.5% 7.5%

Both receive identical recommendations:

  • Same 10-year risk (7.5%)
  • Same statin recommendation
  • Same LDL target (<100 mg/dL)
  • No differentiation

Different strategies based on trajectories:

Woman A (Inflammatory):

  • Earlier intervention (age 50 vs 55)
  • Inflammatory biomarker monitoring
  • Consider imaging (CAC score)
  • Possibly anti-inflammatory Rx

Woman B (Metabolic):

  • Focus on metabolic control
  • Intensive lifestyle intervention
  • Early GLP-1/SGLT2i consideration
  • Tighter glucose monitoring

From Theory to Practice

Precision Prevention by Persona

Current Guidelines:

  • LDL threshold ≥190 mg/dL (FH)
  • Age-independent statin for FH

Life Course Approach:

  • Genetic testing age 30-40
  • PRS-guided intervention
  • Earlier statin initiation
  • More aggressive LDL targets

Evidence:

  • AHA Statement: PRS highly predictive in younger populations
  • Ference et al.: Lifetime cholesterol lowering > short-term intensive

Implementation:

  • Cost: $50-200 for PRS
  • Coverage: Expanding but limited
  • Clinical workflow: Needs EHR integration

Current Approach:

  • Wait for diabetes (A1c ≥6.5%)
  • Reactive treatment

Life Course Approach:

  • Intervene at transition point
  • When metabolic sig plateaus
  • Before CVD sig accelerates

Your Data:

  • Transition occurs 2-3 years pre-CAD
  • Window for intensive prevention

Interventions:

  • Intensive lifestyle (DPP)
  • Early SGLT2i/GLP1-RA
  • Closer monitoring (3-6 months)

Current Gap:

  • Normal lipids = missed
  • Low PCE despite inflammation

Life Course Approach:

  • Inflammatory biomarkers (hsCRP, IL-6)
  • Imaging (CAC) for risk refinement
  • Earlier intervention in rheumatologic disease

Your Finding:

  • Signature deviations 5 years pre-CAD
  • 52 patients with rheum→CV transitions

Treatment Considerations:

  • Optimize rheumatologic disease control
  • Consider colchicine
  • More aggressive lipid lowering
  • Serial imaging

Learn from Success:

  • What protects this group?
  • Can we replicate their trajectory?

Research Needed:

  • Genomic studies
  • Lifestyle patterns
  • Biomarker profiles
  • Microbiome?

Clinical Application:

  • Risk stratification
  • Targeted prevention
  • Personalized goals

Implementation Challenges

Clinical Barriers:

  • EHR integration
    • No life course calculators
    • Data fragmentation
    • Manual entry burden
  • Decision support
    • Need real-time risk updates
    • Trajectory visualization
    • Actionable recommendations
  • Workflow integration
    • Time constraints
    • Training needed
    • Change management

Payer Barriers:

  • Coverage policies
    • “Preventive” Rx in “healthy” patients
    • Genetic testing reimbursement
    • Advanced imaging (CAC, CCTA)
  • Value demonstration
    • Long-term outcomes
    • Cost-effectiveness
    • ROI uncertainty
  • Prior authorization
    • Treatment thresholds
    • Documentation burden

Communication Challenges:

  • Patient understanding
    • Lifetime vs 10-year risk
    • Abstract concepts
    • Motivation for prevention
  • Shared decision-making
    • Risk/benefit discussions
    • Treatment intensity
    • Monitoring frequency
  • Tools needed
    • Visual aids
    • Personalized reports
    • Decision aids

Key Question for Discussion

What’s the threshold for payer coverage of preventive therapies in otherwise “healthy” high-lifetime-risk patients?

Looking Forward

Tools on the Horizon

Current State:

LIVE-CVD Model:

  • First lifelong benefit calculator
  • Integrates traditional risk factors
  • Projects lifetime ASCVD risk

SCOT-HEART Trial:

  • 5-year: CT improves risk stratification
  • 10-year: Sustained benefit
  • SCOT-HEART 2: Ongoing

Your Model (Aladynoulli):

  • Bayesian framework
  • Genetics + biomarkers + events
  • Dynamic, continuous updates
  • 348 diseases, 21 signatures

Emerging Tools:

AI-Enhanced Imaging:

  • Automated plaque quantification
  • Pericoronary fat analysis
  • ECG-based AI risk scores
  • Retinal imaging

Polygenic Risk Scores:

  • Million-variant PRS
  • Multi-ancestry validation
  • Clinical implementation starting

EHR Integration:

  • Real-time calculators
  • Passive monitoring
  • Alert systems
  • Population health dashboards

Wearables & Sensors:

  • Continuous BP monitoring
  • Glucose sensors (CGM)
  • Activity tracking
  • Sleep quality

The Promise: Precision Prevention

The Goal

Tailor prevention to individual life course trajectories—not one-size-fits-all based on 10-year risk alone.

Key Discussion Questions

  1. How do we reconcile population-level risk models with individualized prediction?
    • Especially when individuals may benefit from early intervention despite low calculated risk
  2. Can life-course risk prediction be practically implemented?
    • What barriers do you foresee to payers covering preventive therapies in “healthy” patients?
  3. How do we bridge lifetime risk reduction with current risk management?
    • Lower thresholds? Earlier age to start therapy? Public health efforts?
  4. Communication strategies:
    • How do we make genetics, AI, and imaging meaningful to patient care?
    • What tools help patients understand their lifetime trajectory?
  5. Testing strategies:
    • Should everyone get CAC or CCTA every 10 years starting at age 40?
    • Who should get genetic testing? When?

Summary

Key Takeaways

The Problem:

  • Traditional models miss young high-lifetime-risk patients
  • One-size-fits-all thresholds
  • No consideration of trajectory

The Solution:

  • Life course risk assessment
  • Multiple pathways to disease
  • Dynamic, updating models
  • Precision prevention

The Evidence:

  • Multiple trajectories exist (personas)
  • Signatures change years before diagnosis
  • Velocity predicts risk (HR=2.8)
  • Synergistic multi-signature effects

The Tools:

  • Bayesian modeling (Aladynoulli)
  • Polygenic risk scores
  • AI-enhanced imaging
  • EHR integration coming

The Challenges:

  • Clinical workflow integration
  • Payer coverage
  • Patient communication
  • Training needed

The Path Forward:

  • Validate tools in diverse populations
  • Demonstrate cost-effectiveness
  • Build clinical decision support
  • Engage all stakeholders

Thank You

Contact:

Sarah Urbut, MD PhD Massachusetts General Hospital surbut@mgh.harvard.edu

Preprint:

Urbut et al., medRxiv 2024 doi: 10.1101/2024.09.29.24314557

Code:

github.com/surbut/aladynoulli2

Key Collaborators:

  • UK Biobank participants
  • Mass General Brigham
  • PRIME-Heart Initiative

Questions for Discussion:

  1. Which personas resonate with your practice?
  2. What’s your biggest barrier to life course prevention?
  3. Which tools are you most excited about?
  4. What would change your practice tomorrow?

Ready for Discussion

Let’s explore how we can move from theory to practice in the critical decades.

Backup Slides

Model Details: Aladynoulli

Mathematical Framework:

Disease probability at time \(t\): \[\pi_{i,d,t} = \kappa \times \sum_{k=1}^K \theta_{i,k,t} \times \phi_{k,d,t}\]

Components:

  • \(\theta_{i,k,t} = \text{softmax}(\lambda_{i,k,t})\) — signature proportions
  • \(\lambda_{i,k,t} \sim GP(\mu_k + G_i\gamma_k, K_\lambda)\) — temporal dynamics with genetics
  • \(\phi_{k,d,t} = \text{sigmoid}(\psi_{k,d} + GP(\mu_\phi, K_\phi))\) — disease probabilities

Innovation:

  • Gaussian processes for smooth temporal evolution
  • Genetic effects on signature trajectories
  • Proper censoring handling
  • Bayesian uncertainty quantification

Validation Results

UK Biobank (Primary):

  • N = 400,000 individuals
  • 348 diseases
  • 21 signatures discovered
  • AUC = 0.898 (ASCVD, baseline)
  • AUC = 0.701 (1-year washout)

Cross-Cohort Validation:

  • Mass General Brigham
  • All of Us Research Program
  • 79% signature concordance

Robustness:

  • Stable across ancestry groups
  • Stable across socioeconomic strata (TDI)
  • Population prevalence matches ONS/NHS

Citations & References

  1. Goff et al. 2013 ACC/AHA Pooled Cohort Equations. Circulation 2014.
  2. Lloyd-Jones et al. Prediction of Lifetime Risk for CVD. Circulation 2006.
  3. Khera et al. Polygenic Prediction of Weight and Obesity Trajectories. Cell 2019.
  4. Grundy et al. 2018 AHA/ACC Cholesterol Guidelines. Circulation 2019.
  5. Ference et al. Effect of Long-Term Exposure to Lower LDL. NEJM 2012.
  6. AHA Scientific Statement on Polygenic Risk Scores, 2022.
  7. Ridker et al. Antiinflammatory Therapy with Canakinumab (CANTOS). NEJM 2017.
  8. Harshfield et al. Association Between Genetic Variation at ACE and Risk of COVID-19. Circulation 2020.
  9. Newby et al. Coronary CT Angiography and 5-Year Risk (SCOT-HEART). NEJM 2018.
  10. Williams et al. Coronary Artery Plaque Characteristics (SCOT-HEART 10-year). NEJM 2024.
  11. Urbut et al. Aladynoulli: Bayesian Survival Model for Disease Trajectories. medRxiv 2024.